Viral video analysis that actually looks at the video

What actually happens to your link

Three passes. The second one is the one nobody else runs.

1. 01

   Retrieve and scan

   We fetch the video file and run a first pass over the whole thing — speech, on-screen text, scene boundaries — building a map of where something worth checking might be. This pass alone is roughly where transcript-only tools stop.

2. 02

   Drill into frames

   The scan pass hands back a list of suspicious windows. We go back into the file, extract real frames inside each window at up to 60 frames per second (the mandatory first-3-seconds hook window is cut at 12), and look at them. That is why a finding can say what was on screen at 00:00:02.480, not what was said around then.

3. 03

   Findings pinned to real frames

   A finding we can trace to a real frame carries that frame's timecode, and evidence frames come back annotated; one we cannot trace is shown without a timecode rather than with a guessed one. Milliseconds are derived from frame indices, never written by the model — the number you scrub to is the frame we looked at.

The first-3-seconds hook audit, across three tracks

The first three seconds are audited on every run, even when they look fine. Most hook advice treats a video as one channel. A hook runs on three at once, and they can disagree — that disagreement is usually the problem.

What the YouTube Creator template checks

Three named checks. Each one produces findings pinned to frames, not adjectives.

1. Y1

   Claim vs on-screen evidence

   For every concrete, visually verifiable claim, we look at the frames where the supporting visual should appear and return a verdict: supported, contradicted, or not shown. A transcript tells you the claim was made. Only frames tell you it was never shown.

2. Y2

   First-3-seconds hook audit

   A fixed, always-on window over the opening three seconds, audited across the spoken, visual and on-screen-text tracks, with a delivered-by-3s verdict. It runs even when the opening looks perfect, so you get a baseline on every video.

3. Y3

   CTA consistency

   When the spoken call to action and the on-screen one disagree — different action, different destination, or far apart in time — or when an on-screen CTA is too small to read, we flag the moment and state both sides.

Transcript-level vs frame-level

A capability difference, not a marketing claim. Both approaches answer different questions. Here is what each can and cannot see.

We are not naming competitors, and transcript analysis is not worthless — reading the transcript is the first pass we run too. The point is that the second pass exists.

What this tool cannot do

Every limit below is enforced in code, not aspirational. Better you read this before spending credits than discover it in a report.

• YouTube video comes down at up to 480p

  We cap the download at 480p on purpose. At 720p even a ten-minute video blows past our size ceiling, so nearly every run would degrade and nobody would get frames. Sometimes YouTube's best match under that cap is 360p, so "up to 480p" is literal. Evidence frames show what was in the shot; to read fine print, upload the source file.

• 30 minutes per run, hard limit

  Both YouTube links and uploads cap at 30 minutes. Longer videos are refused outright rather than silently truncated. Trim to the section you care about and upload that.

• Uploads are capped at 50MB

  MP4, MOV and WebM up to 50MB — a smaller ceiling than the YouTube path's 256MB, because an upload holds the whole file in our request path while a YouTube fetch streams to disk.

• Very large videos degrade to scan-level

  If a YouTube video exceeds our 256MB download ceiling — in practice, high-motion footage near the 30-minute limit — the run falls back to a scan-level report with no evidence frames, billed at the lower scan-only rate rather than the standard rate. The link check tells you when we expect that, before you spend anything.

• Some links we simply cannot open

  Private, members-only, age-restricted, region-locked, removed and live videos cannot be retrieved. Platform-side bot checks can also block a fetch that worked yesterday. The MP4 upload path exists because this class of failure is outside our control.

• We do not predict views

  There is no virality score and no forecast of how a video will perform — nobody can honestly derive that from the file alone. You get a list of specific, timecoded things that are working or broken in the edit.

Questions

What is viral video analysis, exactly?

Taking a video that performed well — yours or someone else's — and working out mechanically what the edit did: how the hook lands, whether the claims are shown, whether the call to action is coherent. Here it is a frame-level pass, so findings are tied to specific moments rather than impressions.

Can I analyze someone else's video?

Yes — paste any public YouTube link. Studying videos that worked is how creators learn structure. We return an analysis to you; we do not republish anyone's video.

Is this free?

No. Analysis costs credits and needs an account. We are not a free-score site: a real run downloads the video, samples frames and runs several model passes over them, which costs real money per video. New accounts start with a credit grant. Checking a link costs nothing.

Why does it need my account before it will analyze anything?

A run consumes credits tied to your balance, and the report belongs to your account afterwards. Checking a link also needs sign-in, since each check starts a real retrieval on our side.

How is this different from tools that paste in a transcript?

A transcript holds what was said. It does not hold the title card, the framing, the on-screen CTA, or whether a claim was ever shown. Those live in the pixels. We run the transcript pass too — then we go back into the file and look at the frames.

Why only 480p from YouTube?

A deliberate trade. Higher resolutions blow past our download ceiling on all but the shortest videos, which would degrade nearly every run to a report with no frames at all. 480p is enough to see what was in the shot. For higher fidelity, upload the source file.

What happens if my link fails?

The link check tells you before you spend anything, and offers the upload path instead. If a submitted run cannot retrieve the video, it still produces a scan-level report, billed at the lower scan-only rate rather than the standard rate.

Do Shorts work?

Yes. Shorts, watch links, youtu.be short links and embed links are all accepted. Very short videos bill at a one-minute minimum.

Can I analyze a TikTok or Instagram link?

Not by link today. Download the video and upload the file — the analysis does not care where the footage came from.

What do I actually get back?

A report page with the three-track hook audit, findings from the claim and CTA checks, annotated evidence frames on a filmstrip, and a millisecond timecode on every finding we could anchor to a physical frame, so you can scrub straight to the moment. A finding we cannot anchor is shown without a timecode rather than with a guessed one.